T 2240 Tweet

Munzner T (2009) A nested process model for visualiza-tion design and validation. IEEE Trans Vis ComputGraph 15(6):921–928

Munzner T, Guimbretiere F, Tasiran S, Zhang L,Zhou Y (2003) TreeJuxtaposer: scalable tree compar-ison using focus+context with guaranteed visibility.Proc SIGGRAPH 2003, ACM Trans Graph 22(3):453–462

Mutzel P, Gutwenger C, Brockenauer R, Fialko S, Klau G,Kruger M, Ziegler T, Naher S, Alberts D, Ambras D,Koch G, Junger M, Buchheim C, Leipert S (1998) Alibrary of algorithms for graph drawing. In: The 6thinternational symposium on graph drawing (GD’98),Montreal. Lecture notes in computer science, vol 1547,pp 456–457

Perego UA (2005) The power of DNA: discovering lostand hidden relationships. How DNA analysis tech-niques are assisting in the great search for our an-cestors. In: World library and information congress:71st IFLA general conference and council, Oslo,pp 1–19

Raitner M (2002) HGV: a library for hierarchies, graphs,and views. In: 10th international symposium on graphdrawing, GD 2002, Irvine, pp 236–243

Rozenblat C, Melanon G, Amiel M, Auber D,Discazeaux C, LHostis A, Langlois P, Larribe S(2006) Worldwide multi-level networks of citiesemerging from air traffic (2000). In: Internationalgeographical union IGU 2006 cities of tomorrow,Santiago de Compostela

Schroeder W, Martin K, Lorensen B (2006) The visu-alization toolkit an object-oriented approach to 3Dgraphics, 4th edn. Kitware, Inc., Clifton Park

Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT,Ramage D, Amin N, Schwikowski B, Ideker T (2003)Cytoscape: a software environment for integratedmodels of biomolecular interaction networks. GenomeRes 13(11):2498–2504

Shneiderman B (1996) The eyes have it: a task bydata type taxonomy for information visualizations. In:VL’96: proceedings of the 1996 IEEE symposium onvisual languages, Boulder, pp 336–344

Tominskia C, Abello J, Schumann H (2009) CGV – aninteractive graph visualization system. Comput Graph33(6):660–678

Wylie B, Baumes J (2009) A unified toolkit for infor-mation and scientific visualization. Vis Data Anal7243(1):72430H. http://titan.sandia.gov/ (visited 18Mar 2010)

Tweet

�Microtext Processing�Topic Modeling in Online Social Media, UserFeatures, and Social Networks for

Twitris: A System for CollectiveSocial Intelligence

Amit Sheth, Ashutosh Jadhav,Pavan Kapanipathi, Chen Lu, Hemant Purohit,Gary Alan Smith, and Wenbo WangOhio Center of Excellence inKnowledge-enabled Computing (Kno.e.sis),Wright State University, Dayton, OH, USA

Synonyms

Citizen sensing; Community evolution; Eventanalysis on social media; Interaction network;People-content-network analysis; Real-timesocial media analysis; Semantic perception;Semantic Social Web; Sentiment-emotion-intentanalysis; Social media analysis; Spatio-temporal-thematic analysis; Web 3.0

Glossary

Citizen Sensing Humans as citizens on theubiquitous Web, acting as sensors and sharingtheir observations and views using mobiledevices and Web 2.0 Services

Citizen-Sensor Network An interconnectednetwork of people who actively observe,report, collect, analyze, and disseminateinformation via text, links to other resources,and various media including audio, images,and videos

Social Media Analytics Social media analyticsis the practice of gathering data from socialmedia websites and analyzing that data to gainnew insights from social media and to makeinformed decisions

Semantic Web Semantic Web is a group ofmethods and technologies to help machinesand humans understand the meaning – or“semantics” – of data on the World Wide Web

Spatio-Temporal-Thematic (STT) AnalysisSocial media analytics taking into accountwhat is being said about an event (theme) andwhere (spatial) and when (temporal) it is beingsaid

Twitris: A System for Collective Social Intelligence 2241 T

T

People-Content-Network Analysis (PCNA)Social media analytics taking into accountsocial media user (People), data sharedon social media websites (Content), andnetwork of social media users (Network)

Sentiment-Emotion-Intent (SEI) ExtractionAnalyzing social media content to extractinsights about social media users’ sentiment(positive, negative, and neutral), emotion(happy, angry, upset, etc.), and the user’sintention (seeking information, sharinginformation, etc.)

Introduction

Well over a billion people have become “citizens”of an Internet- or Web-enabled social commu-nity. Web 2.0 fostered the open environment andapplications for tagging, blogging, wikis, andsocial networking sites that have made infor-mation consumption, production, and sharing soincredibly easy. With over five billion mobileconnections, over a billion with data connections(smartphones) and with many more having theability to communicate using SMS, digital mediacan be shared with the rest of the humanityinstantly. As a result, humanity is interconnectedas never before. This interconnected network ofpeople who actively observe, report, collect, ana-lyze, and disseminate information via text, audio,or video messages, increasingly through perva-sively connected mobile devices, has led to whatwe term citizen sensing (Sheth 2009a, b). Thisphenomenon is different from the traditional cen-tralized information dissemination and consump-tion environments where citizens primarily act asconsumers of reported information from severalauthoritative sources.

This citizen sensing is complemented by thegrowing ability to access, integrate, dissect,and analyze individual and collective thinkingof humanity, giving us a capability that isrecognized as collective intelligence. Citizensensing involves humans in the loop and withit all the complexities associated with andintelligence captured in human communication.As citizen sensing has gained momentum, it’s

Twitris: A System for Collective Social Intelligence,Fig. 1 Twitris – three primary dimensions of analysis

generating millions of observations and creatingsignificant information overload. In many casesit becomes nearly impossible to make sense ofthe information around a topic of interest. Giventhis data deluge, analyzing the numerous socialsignals can be extremely challenging. In responseto this growing citizen sensing data deluge,Twitris has been developed with the vision ofperforming semantics-empowered analysis of abroad variety of social media exchanges.

Twitris, named by combining Twitter withTetris, a tile-matching puzzle game, has incorpo-rated increasingly sophisticated analysis of socialdata and associated metadata, combining it withbackground knowledge, and more recently (albeitnot discussed here) machine sensor or data cap-tured from sensors and devices that make up theInternet of Things (IoT). Twitris’ evolution can becharacterized in three phases (and correspondingversions of the system). Figure 1 outlines thecorresponding dimensions Twitris considers.

Twitris is a comprehensive platform for ana-lyzing social content along multiple dimensionsleading to in-depth insights into various aspectsof an event or a situation. The central thesisbehind this work is that citizen sensor obser-vations are inherently multidimensional in na-ture, and taking these dimensions into accountwhile processing, aggregating, connecting, andvisualizing data will provide useful organizationand consumption principles. Twitris evolved in

T 2242 Twitris: A System for Collective Social Intelligence

three phases, characterized by the versions of thesystems:• Twitris v1: Spatio-temporal-thematic (STT)

processing of Twitter and associated news,multimedia, and Wikipedia content (Sheth2009b; Nagarajan et al. 2009a; Jadhav et al.2010)

• Twitris v2: People-content-network analysis(PCNA) (Purohit et al. 2011) with use ofbackground knowledge and semantic meta-data extraction and querying/exploration

• Twitris v3: Sentiment-emotion-intent (SEI)extraction (Chen et al. 2012; Wang et al.2012; Nagarajan et al. 2009b) along withpersonalization (Kapanipathi et al. 2011a) andemerging continuous semantics (Sheth et al.2010) capability involving semantic streamingsocial stream (i.e., real-time) processing usingdynamically generated and updated domainmodels for semantics and contextThe above versions, or phases, of Twitris’

development are not as granular as painted above,that is, the issues identified above are not explic-itly segregated by the version of the Twitris whichhas been in continuous development with seniorstudents graduating and new students picking upthe work. Four talks including a tutorial covermany of the issues covered by Twitris (Sheth2009a; Nagarajan 2010; Nagarajan et al. 2011;Sheth 2011).

Key Points

Throughout this article we investigate the roleand benefits of using semantic approach, espe-cially by metadata extraction and enrichmentsand contextually applying relevant backgroundknowledge, along with demonstrating exampleson real-world data using system (Twitris) de-veloped at Kno.e.sis. The article focuses on thefollowing key points:1. Event-specific analysis of citizen sensing and

discussion on opportunities and challenges inunderstanding temporal, spatial, and thematiccues.

2. Facets of people-content-network analysiswith focus on user-community engagementanalysis.

3. Real-time social media data analysis and theconcept of continuous semantics supported bydynamic model creation.

4. Sentiment and emotion identification from cit-izen sensing data.

5. Recent advances in developing semantic ab-stracts or semantic perception to convert mas-sive amounts of raw observational data intonuggets of information and insights that canaid in human decision making.

Historical Background

The idea of research leading to Twitris occurredon November 26, 2008. Terrorists struck Mum-bai, India, and over the next 3 days, they pro-ceeded to make mayhem in nine locations. Eachof the nine sub-events of this overall event sep-arated by time and location (space) had distinctthematic elements or topical content. The impor-tance of Twitter, especially in terms of citizensensing – the ability of a regular person to usehis or her mobile device to share his or herpersonal observation, thoughts, and belief, wellbefore a traditional news media has a chance to doreporting and to shape opinions – was extensivelydiscussed in the immediate aftermath of this mo-mentous event. This event also gave us a clearcase for the needs and benefits of analyzing socialmedia content such as tweets and Flickr posts andrelated news stories along the three dimensions ofspatial (location of observation), where; temporal(time of observation), when; and thematic (theevent in question), what (Battle 2009; Impact Lab2008; Keralaravind 2008).

Twitris Platform and Three Stages ofIts Evolution

Twitris v1: Spatio-Temporal-Thematic(STT) Processing of Twitter andAssociated News, Multimedia, andWikipedia ContentTwitris v1 (Jadhav et al. 2010) was designed withthe following three major steps:1. Data collection: collect user posted tweets

pertaining to an event from Twitter,

Twitris: A System for Collective Social Intelligence 2243 T

T

Twitris: A System for Collective Social Intelligence,Fig. 2 A snapshot of spatio-temporal-thematic slice ofcitizen sensing showing content related to Mumbai ter-

rorism (thematic) related to Taj hotel (spatial, thematic),during a period of interest (temporal)

associated news, multimedia, and Wikipediacontent (Fig. 2).

2. Data analysis: (a) process obtained tweets toextract strong event descriptors consideringspatial, temporal, and thematic event attributesand (b) process event related news, multime-dia, and Wikipedia content to get event contextand gain a better understanding.

3. Visualization: present extracted summaries onTwitris v1 user interface.Twitris v1 performs a two-step processing to

extract strong event descriptors from tweets.First, it creates the spatio-temporal clustersof the tweet corpus surrounding an event,since every event is different and we want topreserve the social perceptions that generatedthis data. TFIDF computation is performedto fetch the n-grams from this set. Thesecond step involves the association of spatial,temporal, and thematic bias to these n-gramsby means of enhancing the weights whilepreserving the contextual relevance of theseevent descriptors to the event. Further detailsof the text-processing algorithm are availablein (Nagarajan et al. 2009a). Twitris v1 userinterface (Fig. 3) facilitates effective browsingof when (temporal/time), where (space/location),

and what (thematic/context) slices of socialperceptions behind an event.

The objective of the Twitris v1 user interfaceis to integrate the results of the data analysis(extracted descriptors and surrounding discus-sions) with emerging visualization paradigms tofacilitate sensemaking. To start browsing, usersare required to select an event. Once the userchooses a theme, the date is set to the earliest dateof recorded observations for an event and the mapis overlaid with markers indicating the spatiallocations from which observations were made onthat date. We call this the spatio-temporal slice(Figs. 4 and 5).

Users can further explore activity in aparticular space by clicking on the overlaymarker. The event descriptors extracted fromobservations in this spatio-temporal setting aredisplayed as an event descriptor cloud. Thespatio-temporal-thematic (STT) scores determinethe size of the descriptor in the tag cloud. In orderto get event context and better understanding ofthe event, we enhanced Twitris, by integratingevent related news, multimedia (images andvideos), and Wikipedia articles. We leveragedexplicit semantic information from DBpedia toidentify relevant news and Wikipedia articles.

T 2244 Twitris: A System for Collective Social Intelligence

mumbai pakistan pres promised

pakistan pres promised

mumbai attacksphotographers capture imagesphotographers capture images

Event descriptors sorted by their enhanced spatio-temporal-thematic scoresEvent descriptors sorted by their TFIDF scores

photographers capturecapture images

images of mumbaiforeign relations perspective

foreign relations perspectiveindia prime ministercountry of india

attacks in mumbai

1.4553 1.0065 1.71851.58531.52951.50801.45101.37581.37581.37581.3293

1.00171.00171.07971.10831.18221.20481.29331.3028

0.95940.94900.87410.87410.87410.79270.7916

1.39981.27921.21651.12611.09861.09861.08391.0280

photographers capturecapture imagesindia prime ministercountry of india

foreign relations

foreign relationsrejected evidence

rejected evidence

rejected evidence provided

evidence provided

evidence provideduk indicating

uk indicatingmumbai attacks in

mumbai attacks in

mumbai attacksimages of mumbaimumbai

attacks in mumbairejected evidence provided

Twitris: A System for Collective Social Intelligence,Fig. 3 STT biased scoring mechanism of Twitris v1 forrelevance and ranking of keyphrases compared to tradi-

tional TFIDF-based ranking: “mumbai” ranked highestbased on TFIDF is far less informative compared to“foreign relations perspectives”

Twitris: A System for Collective Social Intelligence, Fig. 4 Early version of Twitris v1 user interfaces for displayingthematic component (using STT biasing) on right (b) based on spatial and temporal selection on left (a)

Twitris: A System for Collective Social Intelligence, Fig. 5 Twitris v1 user interface with spatio-temporal slice andmultimedia widgets

Twitris: A System for Collective Social Intelligence 2245 T

T

Twitris: A System for Collective Social Intelligence,Fig. 6 Twitris v1 user interface with event descriptorcloud, related tweets, news, and Wikipedia articles forevent “Austin plane attack.” Joe Stack, the man respon-sible for the Austin suicide plane attack on the IRS office,

put up his suicide note online about the attack. He wasa former bass player for the Billy Eli band. Here Twitriscaptures STT event descriptors summarizing the importantfacets

When a user clicks on a particular descriptor, wedisplay tweets containing the event descriptorsand the top current news items as well as relatedWikipedia articles (Fig. 6).

Twitris v2: People-Content-NetworkAnalysis (PCNA) with Use of BackgroundKnowledge and Semantic MetadataExtraction and Querying/ExplorationThe Mumbai terrorism event of 2008 gavethe impetus to study the event from STTdimensions and focus on connecting withrelevant news content. Social media continuesto grow and revolutionize the way users interactwith each other and information. Social networkusers are not only creators and recipientsof the information but also critical relays topropagate information. This powerful abilityof sharing has played an important role inevents with varied social significance, audience,and duration, such as political movements

(e.g., the Jasmine Revolution in Tunisia), brandmanagement and marketing, and perhaps mostvisibly, crisis and disaster management (e.g.,Haitian and Japanese earthquakes). The Twitristeam started to look at the issues such asthe role of content nature for high vs. lowattributed information diffusion (a phenomenonof propagating messages via friendship/followerconnections among users of social network)(Nagarajan et al. 2010b) and user engagement(given a discussion topic on social media, whatmotivates a user to engage in the discussion forhis/her first interaction) (Purohit et al. 2011;Ruan et al. 2012). Consequently, Twitris v2embarked on a more comprehensive analysisalong the three pillars of what makes anythingsocial: who is engaging in the social activity,what is being communicated, and how does thiscommunication flow between those engaged inthe social activity. The idea is to gain insightsinto how permanent and transient networks arise

T 2246 Twitris: A System for Collective Social Intelligence

Twitris: A System for Collective Social Intelligence, Fig. 7 Contrast in the community structure of influencers inuser interaction networks, centered on two popular events #OccupyChicago and #OccupyLA

Pos

Neg

Objective

Twitris: A System for Collective Social Intelligence,Fig. 8 Sentiment of the influencers for the target can-didate in the interaction network centered on that target:Romney (1st cluster) vs. Ron Paul (2nd cluster)

and what and why information flows across suchnetworks. Twitris v2 developed the significantcapability to extract more types of metadata,and the infrastructure became more semanticwith the use of Semantic Web standard RDFas well as relevant background knowledge. Thelatter enabled Twitris v2 to support the deepexploration capability with use of DBpediaand SPARQL over metadata extracted fromthe tweets. Twitris v2 research that focus oncoordination during disasters also led to integrateTwitris with Ushahidi’s SwiftRiver open sourceplatform and support ingestion of SMS whichwere used for events such as Pakistan Floods in2010.

Let’s look at some examples of Twitris v2capabilities:• Evolving ad hoc nature of social media

communities:Event-centric communities with varied nature(Purohit et al. 2011) often bring together users

from different parts of the social network,especially in Twitter where we keep switchingdiscussions of our interests, and we may notalready be connected to other participantsof those communities. Therefore, in such adhoc communities, it is difficult to depend onjust follower graphs for understanding thedynamics. Twitris v2 introduced analysis of userinteraction networks so that human dynamics inthe evolving communities can be understood atgranular levels – influencer analysis, contextuallyimportant people with roles to engage with,community evolution, etc. Twitris v2 built thisfeature by extending our research in the userinteraction network analysis on brand-pagecommunities (Purohit et al. 2012a).• Contrast in the structure of interaction

networks:Figure 7 shows the networks of influencers intwo topical communities during the Occupy WallStreet (OWS) movement, “OccupyChicago” onthe left and “OccupyLA” on the right. Such ananalysis provides insights to understand not onlythe real dynamics of the actors (e.g., what orga-nizations supporters belong to and to whom arethey strongly connected) but also the potential ofthe influencers to drive actions in the communi-ties (tightly connected influencers are likely todrive effective “call for action” propagation inthe communities). In this figure, the influencernetwork of OccupyLA is highly connected andself-organized as compared to sparsely connected

Twitris: A System for Collective Social Intelligence 2247 T

T

Twitris: A System for Collective Social Intelligence, Fig. 9 Interaction network evolution for topical communitysurrounding Mitt Romney, US Presidential Election 2012

one for OccupyChicago and, therefore, likely toreach masses effectively for any call for action.Even the Facebook page for OccupyLA reflectedsuch activism.• Slicing and dicing the networks by user

features:To glean insights about actionable information inthe ad hoc communities, we need to understandthe participants better. Therefore, Twitris v2introduced slicing and dicing analysis of theinteraction networks by providing user/nodecentric features. For example, professional ororganizational affiliation of users provides cluesto understand the cause for dynamics – e.g., whoare the people behind the organized network ofOccupyLA? Are such users from the same typeof organizations lead to coordinated actions?Similarly, Twitris v2 introduced the content-centric analysis, thus realizing the full potentialof PCNA. Users are clustered by grouping theminto sentiment segments of the target topic, thusanswering questions like which candidate isgoing stronger in the influencer network froma sentiment perspective (Fig. 9) between MittRomney and Ron Paul and for what issues?• Understanding group dynamics by commu-

nity evolution:Twitris v2 focused on the larger goal of predictiveability for group dynamics, and the people-content-network analysis (PCNA) frameworkwas the key to the untapped potential ofgroup dynamics. Therefore, Twitris v2 createdclusters in the ad hoc communities basedon the sentiment of the users for a targeted

topic over time and associated events on thetimeline for causal analytics. Figure 9 showsan example of community evolution centeredaround Republican presidential nominee MittRomney during March 1–31, 2012. It showsthree snapshots taken over a 10-day period,and we observed an extremely modularizedcommunity in the end of the analysis, which wasnot really the case for the closest competition,Rick Santorum. And as we know, Santorumexited the race on April 9. Thus, the analysis ofcommunity evolution made Twitris v2 capableof understanding group dynamics of ad hoccommunities by not limiting the output to justunderstand users but also the group behavior.

Twitris v2 leverages Semantic Web technolo-gies by the use of background knowledge suchas DBpedia to provide deeper insights about theevent. Background knowledge changes the wayyou can look at the information, as it puts the in-formation in context. This is especially importantfor tweets because they are short and thereforeindividually lack the volume of information thatprovides an informative context. For example,in the above Fig. 10, questions such as “Whoare the dead people that are mentioned in thecontext of OWS movement” can be answeredusing the background knowledge, whereas simplekeyword search cannot put the information oftweets in context. Further, to answer the questionsin the figure and generate answers such as RosaParks, the system has to have the backgroundknowledge about this named entity as a Personand also that she is dead. Going deeper into the

T 2248 Twitris: A System for Collective Social Intelligence

Twitris: A System forCollective SocialIntelligence, Fig. 10Leveraging Semantic Webtechnologies to provideinsights of events

background knowledge provides information thatRosa Parks was famous for the Montgomery BusBoycott during the US civil rights movement in1955–1956.

Twitris v3: Emotion-Sentiment-Intent,Real-Time View and Other AdvancementsBehind every (well, most of the important) tweet,there is a human. And a human is complex.Through a tweet, a person expresses emotion,sentiment, and intent. Understanding this dimen-sion is a key to unlock the true potential of socialmedia. This is especially true for monetizationof social media. Understanding an underlyingintent can tell us if a user is expressing a trans-actional (potentially for buying a product) intent,seeking information, or just sharing information(Nagarajan et al. 2009b). Sentiment is perhaps themost sought after type of analysis of social data.Currently, it is the primary basis of social media

analysis to predict whether a product or a moviewill succeed, who is more likely to win an elec-tion, or to attempt to identify consumer interestand hence use it for targeting the advertisement.Analysis of or identification of emotion is likelythe dark horse of the three – while techniques forits analysis are not yet as mature as sentimentanalysis, it is likely to be combined with the othertwo to give far more signal than without it.

A key innovation in sentiment analysis,employed in Twitris v3, is topic- specificsentiment analysis – to associate sentiment withan entity (Chen et al. 2012). This enables us toidentify two different sentiments associated withdifferent entities in a single tweet. For example,in tweet “The King’s Speech was bloody brilliant.Colin Firth and Geoffrey Rush were fantastic!”we can identify both the sentiment (i.e., bloodybrilliant) associated with the movie “The King’sSpeech” and the sentiment (fantastic) associated

Twitris: A System for Collective Social Intelligence 2249 T

T

Twitris: A System for Collective Social Intelligence, Fig. 11 Twitter users show the opposite sentiments towardstwo candidates on the same topic “final debate” in 2012 presidential election

with the actors Colin Firth and Geoffrey Rush.More recently, we are associating sentimentswith events – when there is a significant changein sentiment, we attempt to associate that withreal-world events. For example, by tracking boththe event- and entity-specific sentiments, Twitrisv3 is able to capture a substantial increase ofpositive sentiment towards President Obamaon the immigration issue on June 15, 2012(the day on which President Obama outlineda new immigration policy), and associate itwith the event descriptors such as “dreamact,” “obama ’s immigration move,” and “new

immigration policy.” Figure 10 shows that Twitterusers have the opposite sentiments towardstwo candidates: Obama (green/positive) andRomney (red/negative) on the same topic “finaldebate.” The reason is that Obama receivedmore positive feedback from Twitter users thanRomney did, which is in line with the impressionfrom news media. This example demonstratesTwitris’s power in identifying topic-specificsentiments.

Compared with sentiment, emotion is moreimplicit. For example, “I will have a calculus testin two hours, but I’m not prepared at all.” We

T 2250 Twitris: A System for Collective Social Intelligence

Twitris: A System for Collective Social Intelligence, Fig. 12 Peek patterns of the emotion joy due to excitement ofTwitter users caused by three debates and one TV program (the Daily Show) in the 2012 presidential election

can infer that the person is nervous about the test,though there are no explicit emotion words, suchas “nervous” or “panic.” It is very difficult andtime consuming to label sentences with emotions,considering the implicitness of emotion. InTwitris v3, we are able to automatically createa large emotion-labeled dataset (of about2.5 million tweets) covering: joy, sadness,anger, love, fear, thankfulness, and surprise, byharnessing emotion-related hashtags available inthe tweets (Wang et al. 2012). Machine learningclassifiers are trained on the large dataset to learnhow to identify people’s emotions behind theirtweets. And, as another key innovation, Twitrisv3 can analyze people’s emotional responsesin different events. For example, Fig. 11 showsthe volume of joyful tweets, reaching peaks onOct. 3, 2012 (first debate), Oct. 16, 2012 (seconddebate), Oct. 22, 2012 (third debated), and Oct.19, 2012 (Obama went to the Daily Show). Thereason is that Twitter users are very enthusiasticabout all three presidential debates and Obama’spresence in the Daily Show TV program. Otherthan analyzing emotions out of tweets, Twitrisv3 is also able to identify emotions from blogs,news headlines, etc. The reason is that we adaptthe classifiers trained on Twitter data to otherdomains with a relatively small amount of labeledemotion data in other domains.

Besides detecting users’ emotional states, wealso explore how to automatically identify users’intents from posts so that monetization can bemore targeted on users’ needs (Nagarajan et al.2009b). The highlight of our study is that wediscover and differentiate three types of posts: (a)transactional posts, e.g., “I am looking for a 32GB iTouch”; (b) information sharing posts, e.g.,“I like my new 32 GB iTouch”; and (c) infor-mation seeking posts, e.g., “what do you thinkabout 32 GB iTouch?” For monetization pur-poses, transactional posts and information seek-ing posts are more valuable than informationsharing posts because users are looking for infor-mation that advertisers can exploit. By extract-ing intent/keywords/cues from transactional andinformation seeking posts, our system achievedan accuracy of 52 % on ad impressions usingMySpace and Facebook data, while the baseline,without using our system, only achieved an accu-racy of 30 %.

All the above-mentioned precious assets (sen-timent, emotion, and intent) of content exist dueto an actionable purpose of humans. When suchindividual level purposes start to bring higherengagement in the groups, they become sourceof group-level actions, apparently, leading to theevolution of human dynamics in the social net-work. Therefore, we are exploring the integral

Twitris: A System for Collective Social Intelligence 2251 T

T

Twitris: A System for Collective Social Intelligence,Fig. 13 Some of the capabilities of Twitris v3: (1)show popular Topics, also called social signals (weightedn-grams) related to the chosen event for today and anyday of the past since the event began to be tracked;(2) search from among the event related tweets withautocomplete, popular event hashtags, and active usersand explore content for deep analysis (e.g., who arethe dead people mentioned most often in the “occupywall street event”) using background knowledge (defaultsource is Wikipedia/DBpedia) and Semantic Web tech-nologies (RDF/SPARQL); (3) show key topics of dis-cussions by locations/regions, states, and country (e.g.,see the differences in social signals from Mississippi (a“red state”) vs. Massachusetts (a “blue state”) related toPresident Obama’s Nobel Prize); (4) see event relevant

tweets in real time on a world map or any region; (5)analyze topic-/people-/region-specific sentiment (e.g., forthe US election, sentiment on candidates by states, andby topics identified by election specific topics); (6) seethe networks with insights from static (e.g., followers) anddynamic features (e.g., retweet) and people/demographics(e.g., with knowledge of profession of each person); (7)display tweets, recent news, and Wikipedia pages relatedto selected events and social signals; (8) show event-specific multimedia (images and video); (9) see tweettraffic; (10) change date of video/analysis; (11) selectlocation of interest – each pin shows a collection of socialsignals emanating from a location; and (13) select anevent of interest (e.g., US Election, Occupy Wall Street,Japanese Tsunami)

role of intent with sentiment and emotions forpurposeful actions in the groups. Specifically,we are focusing on intent and sentiments behindgroup coordination because coordinated activityhas the potential to make or break the system(Fig. 13).

Advanced ResearchDetailed research on social data analysis encom-passes social intelligence in real time (Gruhlet al. 2010) which involved a Kno.e.sis-IBM

collaboration leading to the operationallydeployed BBC Sound Index system, predictionof topic volume on Twitter (Ruan et al.2012), emotion identification using Twitter“big data” (Wang et al. 2012), brand tracking(Purohit et al. 2012a), psycholinguistic analysisduring emerging coordination (Purohit et al.2012b), privacy-aware content dissemination(Kapanipathi et al. 2011b), user-communityengagement (Purohit et al. 2011), informationdiffusion (Nagarajan et al. 2010b), trust in social

T 2252 Twitris: A System for Collective Social Intelligence

media (Thirunarayan and Anantharam 2011),and monetization of social activities (Nagarajanet al. 2009b) reported in over 30 publicationsand summarized in a comprehensive tutorial(Nagarajan et al. 2011).

Key Applications

Twitris has been used in a research context forstudying and analyzing social sensing and per-ception of a broad variety of events: politics andelections, social movements and uprisings, crisisand disasters, entertainment, environment, etc.We are now investigating more commercial ap-plications including brand tracking and advertise-ment campaign effectiveness, and empoweringprofessional users.

Future Directions

The next evolution of Twitris will be in incor-porating social media content along with datafrom sensors and Web of Things, as well asin advanced applications for crisis and disastersupport.

Acknowledgments

Alumni, Karthik Gomadam, Meena Nagarajan,and Ajith Ranabahu, have made substantialcontributions to Twitris. Twitris continues tobenefit from current collaborators, includingProf. Krishnaprasad Thirunarayan and Prof.Valerie Shalin, and team members PramodAnantharam and Shreyansh Bhatt. This workis partially supported by NSF funded grants“SoCS: Social Media Enhanced OrganizationalSensemaking in Emergency Response” (IIS-1111182) and “EAGER: Expressive ScalableQueries over Linked Open Data” (IIS-1143717).

Cross-References

�Collective Intelligence, Overview�Dynamic Community Detection

�Futures of Social Networks: Where Are TrendsHeading?�Semantic Social Networks Analysis�Spatiotemporal Footprints in SocialNetworks�User Sentiment and Opinion Analysis

References

Battle C (2009) New media’s moment in Mumbai.Foreign Policy J, 15 Jan 2009. http://www.foreignpolicyjournal.com/2009/01/15/new-media. Accessed31 Mar 2013

Chen L, Wang W, Nagarajan M, Wang S, Sheth A (2012)Extracting diverse sentiment expressions with target-dependent polarity from Twitter. In: Proceedings ofthe 6th international AAAI conference on weblogs andsocial media (ICWSM), Dublin, 5–7 June 2012

Gruhl D, Nagarajan M, Pieper J, Robson C, Sheth A(2010) Multimodal social intelligence in a real-timedashboard system. VLDB J (Data Manage Mining SocNetw Soc Media) (Special issue, to appear) 19(6): 825-848

Impact Lab (2008) Twitter provided a vital link in Mumbaiterrorist attacks, November 28, 2008. http://www.impactlab.net/2008/11/28/twitter-provided-a-vital-link-in-mumbai-terrorist-attacks/. Accessed 31 Mar 2013

Jadhav A, Purohit H, Kapanipathi P, Ananthram P, Ran-abahu A, Nguyen V, Mendes P, Smith AG, CooneyM, Sheth A (2010) Twitris 2.0: semantically empow-ered system for understanding perceptions from socialdata. In: Semantic web application challenge at ISWC,Shanghai, 7–11 Nov 2010

Kapanipathi P, Orlandi F, Sheth A, Passant A (2011a)Personalized filtering of the Twitter stream. In: 2ndworkshop on semantic personalized information man-agement at ISWC 2011, Koblenz, 23–27 Oct 2011

Kapanipathi P, Anaya J, Sheth A, Slatkin B, Passant A(2011b) Privacy-aware and scalable content dissemi-nation in distributed social networks. In: Internationalsemantic web conference (ISWC), Koblenz, 23–27Oct 2011

Keralaravind (2008) ‘Hash Mumbai’, time line of citizenjournalism & social media during Mumbai terrorist at-tacks, Youtube.com, uploaded on 28 Nov 2008. http://www.youtube.com/watch?v=copw-W-IfvY. Accessed31 Mar 2013

Nagarajan M (2010) Understanding user-generated con-tent on social media. Ph.D. dissertation, Wright StateUniversity

Nagarajan M, Gomadam K, Sheth A, Ranabahu A,Mutharaju R, Jadhav A (2009a) Spatio-temporal-thematic analysis of citizen-sensor data – challengesand experiences. In: Tenth international conferenceon web information systems engineering, Poznan, 5–7Oct 2009